No agent is installed on a PLC, and most field devices don’t produce logs. For attack detection on an OT network, the most valuable signal you have is the industrial protocol commands passing over the wire. The endpoint- and log-centered monitoring habit from IT doesn’t work here; the logic has to be built from scratch. In this guide we explain sensor placement, the normal communication map, detection of PLC program and mode changes, and how these alarms are passed to the SOC. The target readers are security teams that want to turn their OT visibility infrastructure into attack detection.
Why OT monitoring differs from IT monitoring
On an IT network, traffic is scattered: users connect to new destinations every day, applications get updated, cloud services change. On an OT network, communication is largely deterministic; the same HMI reads the same registers from the same PLC at the same period. That’s both an advantage and a limitation for monitoring.
- Advantage: Because normal behavior is defined in a narrow range, deviation stands out. A new IP connecting to a PLC is noise in IT; in OT it’s a meaningful signal.
- Limitation: The attacker’s commands are the same as legitimate engineering commands. Loading a program onto a PLC isn’t malicious on its own; what matters is who loaded it, when and with which change record.
- Response constraint: In IT, a suspicious connection can be cut automatically. In OT, a wrong automatic block cuts process control traffic and the line stops. That’s why OT monitoring mostly works on a model of passive detection and response through human decision.
Sensor placement and visibility plan
Detection is only as good as the traffic the sensor sees. The collection points set up for asset discovery are a good start; the SPAN/TAP placement in the OT passive asset discovery guide applies here too. For attack detection, these flows also need to be visible:
- The IT/OT boundary and the OT DMZ: Remote access, historian replication, file transfer. Attacks mostly enter OT from here.
- Between the engineering workstation and the PLC: Program upload, firmware update and mode change commands pass through this path.
- Between SCADA/HMI and the PLC: Writing setpoints and sending commands.
- Intra-cell traffic (on critical lines): An unauthorized laptop plugged directly into the cell switch is visible only here.
Put the visibility plan in a table; the blind spots reveal themselves. Three columns per cell are enough: “is supervisory traffic visible”, “is intra-cell traffic visible”, “through which point does engineering traffic pass”. What isn’t visible is recorded as a conscious risk decision. An unrecorded blind spot is a forgotten blind spot.
Baseline: building a normal communication map
The foundation of behavior-based detection is a model of normal communication. Most commercial OT monitoring platforms use a learning period; on the open-source side, you can build the same map from Zeek logs. The baseline keeps these dimensions:
- Source–destination pairs and the protocol/port used.
- Operation type at the protocol level: read or write, which function codes or CIP services.
- Period and volume: how often the HMI polls the PLC, the typical packet count.
- Time pattern: on which days and in which windows engineering traffic occurs.
Two mistakes come up again and again in the learning period. The first is the period covering only routine production; the first recipe change or planned maintenance that follows produces hundreds of alarms, and the team loses trust in the sensor. The second is assuming the network is already clean. When learning ends, the baseline should be read line by line with the operations and automation team. The question “why does this laptop connect to the PLC every night?” is usually asked in this meeting.
Critical events: PLC writes, firmware and mode changes
Regardless of the anomaly model, some events should raise an alarm in every case. These are commands that can directly affect the process:
| Protocol | Operation to monitor | Why it’s critical |
|---|---|---|
| Modbus TCP | FC 5, 6, 15, 16 (write), FC 22–23; FC 8 subfunction 4 (Force Listen Only) | Changes outputs and registers; FC 8/4 can knock the device off communication |
| S7comm | Download (0x1A–0x1C), PLC Stop (0x29), PI services (0x28), block delete | Changes the program, stops or starts the CPU |
| EtherNet/IP / CIP | Reset service on the Identity object, Set_Attribute requests, firmware upload | Restarts the device or changes its configuration |
| DNP3 | Select/Operate, Direct Operate, Cold/Warm Restart (FC 13/14) | Drives field outputs or restarts the device |
The value of these alarms comes with context: is the source sending the command an approved engineering workstation, does the time fall within a maintenance window, is there an open change record for this PLC? An alarm without context is nothing more than an IP address landing in front of the analyst. With the S7comm-plus protocol used by newer Siemens S7-1200/1500 CPUs and with encrypted sessions, content decoding is limited; in that case, detection at the level of “an engineering session was opened” is used together with periodic checks of the CPU diagnostic buffer.
On the open-source side, Suricata offers application-layer keywords for Modbus. The example rule below raises an alarm on Modbus write operations coming from outside the approved engineering workstations:
# suricata.yaml: app-layer.protocols.modbus.enabled must be yes
# address-groups: EWS_HOSTS and PLC_NET must be defined
alert modbus !$EWS_HOSTS any -> $PLC_NET 502 (msg:"OT - Modbus write from unauthorized source"; \
modbus: access write; classtype:policy-violation; sid:9000101; rev:1;)
alert modbus any any -> $PLC_NET 502 (msg:"OT - Modbus FC8 diagnostic request"; \
modbus: function 8; classtype:attempted-dos; sid:9000102; rev:1;)
A common setup mistake is this: the Modbus parser is disabled in Suricata’s default configuration, the rule is loaded but never matches. A silent rule gives a sense of security, not protection. Before putting rules live, try them on recorded traffic (pcap) with suricata -r; you’ll see both the matches and the noise in advance.
The limits of signature- and anomaly-based detection
| Approach | Strength | Limitation |
|---|---|---|
| Signature-based | Clearly recognizes known malicious tools, exploits and IT-originated threats | Doesn’t see attacks carried out with legitimate commands; the number of OT-specific signatures is limited |
| Policy/rule-based | Definite rules like “this source can’t write to this PLC”, little noise | Writing rules requires a good inventory and process knowledge |
| Behavior/anomaly | Catches unknown methods as deviations | Noise on unplanned but legitimate changes; depends on baseline quality |
| Process value monitoring | Notices physically impossible or dangerous setpoints | Requires process knowledge and joint work with the automation team |
What works in industrial cybersecurity monitoring is layering these approaches. IT-originated threats (ransomware spread, remote access tools, credential attacks) are caught at the OT boundary with signature and behavior detection. OT-specific manipulation is caught with policy rules and the critical event list. A single layer isn’t enough. Marking detection coverage on MITRE’s ATT&CK for ICS matrix shows concretely which techniques can’t be seen at all.
SOC integration and triage of OT alarms
The OT sensor raises the alarm; the SOC analyst who will evaluate that alarm on the night shift often doesn’t know what the PLC does. Integration has three components:
- Enrichment: The alarm should arrive in the SIEM with the asset name, line/cell, criticality and responsible person. An OT alarm arriving with only an IP address lengthens triage.
- Change record correlation: If a PLC write or program upload alarm matches an approved change record or maintenance work order in the same time window, its priority drops; if it doesn’t, it rises.
- Escalation path: The SOC shouldn’t have the authority to intervene directly in OT. The analyst’s job is to validate the alarm and pass it to the shift engineer or OT security owner through a defined scenario.
Example triage flow: at 2:40 a.m., an S7 download to a PLC is detected. The analyst looks at the source first; if the source isn’t on the approved engineering workstation list, the event goes to high priority. Then the change record is checked; if there’s no record, the shift engineer is called and asked to check the line’s status. If the engineer doesn’t recognize the change, the event is handed over to the isolation decision process in the OT incident response plan. Every step of this flow should be written in the SOC playbook. If it isn’t, at 2:40 a.m. nobody knows what to do.
For organizations planning to connect OT monitoring to their existing SOC service, SOC and MDR consulting is the starting point, and for sensor design and the rule set, the OT/ICS security service. For protocol-level blocking rules on the firewall side, the Modbus and S7 deep packet inspection guide on the industrial site is a complement.
What to measure
- Visibility coverage: the ratio of cells monitored by a sensor to the total number of cells.
- How critical event alarms match change records: every unmatched alarm is either a process gap or a security gap.
- Triage time: the time from the alarm being raised to feedback from the shift engineer.
- Periodic testing: in a maintenance window, a harmless read/write attempt is made from an unapproved station, and it’s verified that the alarm reaches the SOC. An untested alarm chain doesn’t count as working.
Frequently asked questions
Should automatic blocking with an IPS be done on the OT network?
Be cautious about automatic blocking where process control traffic passes; a single false match can stop production. The common approach: block IT-originated threats at the OT boundary, and inside the cell detect and respond through human decision.
How long should the baseline learning period be?
Coverage matters more than duration. The learning period should include shift changes, product or recipe changes and at least one planned maintenance. After that, the learned relationships should be reviewed one by one with the operations team.
Is detection possible with encrypted PLC protocols?
Content-level detection narrows; but the source of the connection, its timing, session establishment and volume changes are still visible. When this information is combined with the PLC diagnostic buffer and project file comparison, unauthorized changes can be caught.
What should a SOC analyst do with an OT alarm?
The analyst validates the alarm against asset information and change records and passes it to the shift engineer or OT security owner through the defined escalation path. The decision to intervene directly in an OT system is made together with the operations team.